Design And Implementation Of Vehicle Detection In Traffic Scene

It is the important premise for intelligent transportation system (ITS) that to gain real-time traffic information. ITS has been improved by video monitor and video analysis technology. Video surveillance which spans many subjects is one of the chief fields in IT technology. The key technologies of intelligent video surveillance include object detection, object segmentation and object tracking, etc.In this paper, we focus on moving object detection and tracking in traffic scenes. Our work includes adaptive background modeling, object segmentation, object tracking and other issues. We propose new algorithms and improving methods to related work, and develop an algorithm framework for real surveillance system. At last, we implement these algorithms in embedded DSP hardware. The main work can be illustrated in the following four parts:Firstly, we do research in moving pixel detection. Many proposed algorithms are based on adaptive background model and some of them achieve good performance. Our algorithm in moving object detection is also based on adaptive background modeling method. It’s named Balloon Model and established in RGB color space. Compared with the existed algorithms, which including the adaptive background models in VSAM, Gaussian Mixture Model and so on. The proposed method achieves good performance in the accuracy and real-time property.Secondly, we focus on the segmentation of moving objects. It’s known that accurate moving object segmentation is an essential problem in intelligent video surveillance system. However, the existence of unexpected moving cast shadows frequently leads to errors in further scene analysis. This paper presents a novel method that combines color space and corner feature to detect and to remove cast shadows of moving vehicles in the traffic sequences. The two features cooperate well to classify vehicles and cast shadows by means of Multiple Masks Method (MMM). The proposed algorithm has been tested on video sequences taken under various shadow orientations and shadow sizes, various vehicle colors and vehicle sizes, different illumination conditions. The results have revealed that shadows can be successfully eliminated and thus good vehicle segmentation can be obtained.Thirdly, we propose a new object tracking algorithm, Curved Sliding Method, which is based on area-based tracking strategies. It origins from Gradient Descent Method and established in discrete space. The result of sliding surface tracking algorithm iterates and chose the local optimization step in sliding surface of matching criteria, which achieves global optimum effects with expectation. These three parts unite together and forms a complete video-based motion detection and analysis system.Lastly, through the study of the method in motion detection, shadow detection and target segmentation, we can put forward an algorithm framework to build an video analysis system. We integrate the algorithms into a real-time video analysis algorithm framework. The algorithm framework is realized in a system software. We implement the embedded DSP hardware to gain traffic flow parameters.